vllm-project/llama-68m

[!NOTE] This is a copy of JackFram/llama-68m at commit 9de8453, made on 2026-10-30. vLLM CI uses this checkpoint and the original is on a personal account, so we keep a copy here to reduce supply-chain risk.

Model description

This is a LLaMA-like model with only 68M parameters trained on Wikipedia and part of the C4-en and C4-realnewslike datasets.

No evaluation has been conducted yet, so use it with care.

The model is mainly developed as a base Small Speculative Model in the SpecInfer paper.

Evaluations (contributed by Akshit, huge thanks!)

Notes on Failed Tasks: The Arithmetic and SocialIQA benchmarks failed during execution due to runtime pipeline incompatibilities, yielding no score. Total evaluation runtime was 44.74 minutes.

Citation

To cite the model, please use

@misc{miao2023specinfer,
      title={SpecInfer: Accelerating Generative LLM Serving with Speculative Inference and Token Tree Verification}, 
      author={Xupeng Miao and Gabriele Oliaro and Zhihao Zhang and Xinhao Cheng and Zeyu Wang and Rae Ying Yee Wong and Zhuoming Chen and Daiyaan Arfeen and Reyna Abhyankar and Zhihao Jia},
      year={2023},
      eprint={2305.09781},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}